For our purposes, inertial navigation is defined by the Inertial Navigation System (INS). The INS is composed of three principle components: the core inertial sensor suite in the IMU, the aiding sensors, and the INS algorithms and communications systems implemented in software. While there are many different implementations of INS technology, the descriptions here are specific to Greensea's INS technology and methodology.
Greensea's INS provides a navigation solution in the ECEF coordinate frame by fusing measurements from available sensors with measurements from a high-rate Inertial Measurement Unit (IMU). Typically we employ this as an aided inertial system in that the inertial measurements provide the core algorithm process using any other available state measurement to aid the inertial solution. In terms of stability and noise, the inertial measurements are actually pretty lousy for long-term accuracy in the navigational solution. Because this is the highest-rate sensor and it measures core body motion, it is very convenient mathematically to have this as the core process in the fusion system.
Greensea's INS fuses aiding state measurements and core inertial measurements with a discretized variant of the basic Kalman filter (KF). This filter uses linear error and observation models to produce state estimations that are more accurate than any individual discrete input measurement alone. The filter algorithms used in the INS are configured to account for the noise, bias, and variable update rates of the IMU and aiding sensors.
To simplify the process, the function of the INS is best visualized by considering the contribution each sensor has to a single final state estimation in a simple example. Consider a USBL. USBL data typically has good long term and trending accuracies but the sample-to-sample error and noise can be quite problematic. Additionally, USBL data has a relatively slow update rate (<1Hz). An IMU on the other hand offers high-rate data (>50Hz) that is highly subject to long-term drift but offers reasonable accuracy on a short time scale. Using stochastic processes, an INS builds an understanding of the errors associated with each sensor. The INS considers the IMU as the principle sensor and builds a solution estimate based on the inertial data. When a USBL update is received, the INS algorithm corrects the solution estimate based on the USBL data while considering the learned error from each data source.
